PufferLib is a reinforcement learning library designed to train tiny, super-human models efficiently. It includes proprietary learning algorithms, hyperparameter tuning, and simulation methods developed through the maintainers’ own research. The library is open source and targets researchers and developers building RL-based applications.